Báo cáo hóa học: "Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival"

Chia sẻ: Linh Ha | Ngày: | Loại File: PDF | Số trang:10

Thêm vào BST

Báo xấu

55
lượt xem 6
download

Download Vui lòng tải xuống để xem tài liệu đầy đủ

Tuyển tập báo cáo các nghiên cứu khoa học quốc tế ngành hóa học dành cho các bạn yêu hóa học tham khảo đề tài: Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival

Chủ đề:

Bình luận(0) Đăng nhập để gửi bình luận!

Lưu

Nội dung Text: Báo cáo hóa học: "Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival"

Tambasco et al. Journal of Translational Medicine 2010, 8:140 http://www.translational-medicine.com/content/8/1/140 RESEARCH Open Access Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival Mauro Tambasco1,2,3*, Misha Eliasziw1,4, Anthony M Magliocco1,2,5 Abstract Background: Precise criteria for optimal patient selection for adjuvant chemotherapy remain controversial and include subjective components such as tumour morphometry (pathological grade). There is a need to replace subjective criteria with objective measurements to improve risk assessment and therapeutic decisions. We assessed the prognostic value of fractal dimension (an objective measure of morphologic complexity) for invasive ductal carcinoma of the breast. Methods: We applied fractal analysis to pan-cytokeratin stained tissue microarray (TMA) cores derived from 379 patients. Patients were categorized according to low (1.75, N = 90) fractal dimension. Cox proportional-hazards regression was used to assess the relationship between disease-specific and overall survival and fractal dimension, tumour size, grade, nodal status, estrogen receptor status, and HER-2/neu status. Results: Patients with higher fractal score had significantly lower disease-specific 10-year survival (25.0%, 56.4%, and 69.4% for high, intermediate, and low fractal dimension, respectively, p < 0.001). Overall 10-year survival showed a similar association. Fractal dimension, nodal status, and grade were the only significant (P < 0.05) independent predictors for both disease-specific and overall survival. Among all of the prognosticators, the fractal dimension hazard ratio for disease-specific survival, 2.6 (95% confidence interval (CI) = 1.4,4.8; P = 0.002), was second only to the slightly higher hazard ratio of 3.1 (95% CI = 1.9,5.1; P < 0.001) for nodal status. As for overall survival, fractal dimension had the highest hazard ratio, 2.7 (95% CI = 1.6,4.7); P < 0.001). Split-sample cross-validation analysis suggests these results are generalizable. Conclusion: Except for nodal status, morphologic complexity of breast epithelium as measured quantitatively by fractal dimension was more strongly and significantly associated with disease-specific and overall survival than standard prognosticators. Background Currently, the most significant prognosticator for The prognostic assessment of breast cancer is based on women with breast cancer is axillary lymph node status factors that determine a patient ’ s relapse risk, and [1-4]. For node-positive patients, there is a direct rela- together with predictive factors (e.g., estrogen-receptor tionship between the number of involved axillary nodes status), it is used to make optimal therapeutic decisions and the risk for distant recurrence [4]. However, despite regarding adjuvant systemic therapy [1]. Such decisions the usefulness of lymph node status, recommendations provide a balance between the potential benefit and for systemic adjuvant chemotherapy are not entirely associated costs and side effects of treatment [1]. There- straightforward. For exampl e, five-year survival rates fore, it is necessary to have sensitive and specific prog- show that approximately 15% of all node-negative nosticators to accurately define risk category for breast patients with larger tumor sizes (>1 cm) may benefit cancer. from systemic adjuvant therapy, but about 85% would survive without it [5]. Furthermore, approximately one- third of node-positive patients are free of recurrence * Correspondence: mtambasc@ucalgary.ca after local-regional therapy [6-8]. 1 Department of Oncology, University of Calgary, Calgary, Canada Full list of author information is available at the end of the article © 2010 Tambasco et al; licensee BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 2 of 10 http://www.translational-medicine.com/content/8/1/140 O ther major prognostic risk factors, especially for tissue for TMA construction, and had received adjuvant node-negative patients, are tumor size and histological tamoxifen treatment but no adjuvant chemotherapy. tumor grade [1-4,9,10]. For node-negative patients, tumor size is a powerful prognostic factor that is used Sample Preparation routinely to make adjuvant treatment decisions [6,11], Whole sections stained with Hemotoxylin and Eosin and tumor grade is primarily used to make decisions for (H&E) were used to select tumor areas for the TMA cases in which the tumor sizes are borderline [1,2,5]. cores. Fourteen breast TMA blocks containing an average Although tumor grade has prognostic value, significant of 94 tissue cores were constructed from formalin-fixed, inter-observer variation in grading still exists [12-14]. as paraffin-embedded, previously untreated breast cancer pathologists are assessing complex histological charac- tissue. To ensure there was no selection bias, three teristics in a semi-quantitative manner. 0.6 mm cores were chosen randomly from cancerous It is known that invasive breast cancer (a malignant areas of each donor block to construct the recipient neoplasm) demonstrates partial or complete lack of TMA core block, and the Leica RM2235 microtome (Leica Microsystems Inc.) was used to cut 4 μ m thick structural organization and functional coordination with surrounding normal tissue [15]. The idea central to this sections from each TMA donor block. In a previous study is that this loss of structural organization and study with prostate cancer specimens, we showed that functional coordination manifests itself in the form of fractal analyses of specimens stained with pan-cytokera- an increase in morphologic complexity of the epithelial tin provide greater classification performance (benign components at the sub-cellular, cellular, and multi-cellu- versus high grade) than serial sections of the same speci- lar levels, and the degree of this complexity can be mens stained with H&E [18]. The reason for this is that quantified and related to patient outcome. A method pan-cytokeratin isolates and highlights the morphology that lends itself particularly useful for quantitatively of epithelial components and excludes structures that do characterizing complex pathological structures at differ- express pathological relevance in the form of morpholo- ent scales, is based on fractal analysis [16,17]. In this gic complexity (i.e., connective tissue components). study, we assess the prognostic value of a recently devel- Hence, we stained all the TMA sections with pan- oped novel technique [18] to measure the fractal dimen- cytokeratin. This staining was performed using Ventana sion of segmented histological structures of breast tissue Benchmark LT. Protease 1 antigen retrieval was used fol- microarray (TMA) cores stained with pan-cytokeratin to lowed by Ventana pre-diluted pan-cytokeratin (cat. No. highlight the morphology of epithelial architecture. 760-2135) antibody with an incubation time of 32 min- utes. A Ventana ultraview™ DAB detection system was Methods used for detection. Patient Characteristics A total of 408 patients with primary invasive ductal car- Image Acquisition of TMA Cores cinoma (IDC) of the breast were selected retrospectively Microscopic images of the TMA cores were acquired from the Calgary Regional Hospitals after appropriate with an AxioCam HR digital camera (Carl Zeiss, Inc.) ethics approval from the Institutional Review Board mounted on an optical microscope (Zeiss Axioscope) at (IRB). It should be noted that the IRB did not require a magnification of 10 × objective. The AxioCam HR has pixels of size 6.7 μ m × 6.7 μ m , which are 1.06 μ m × patient consent for this study as it was a retrospective 1.06 μm in apparent size at the combined magnifications study in which many of the patients were deceased and the risk of exposing patient confidentiality was extre- of 10 × objective and 0.63 × C-mount optical coupling mely low. Of these, 379 patients had at least one of (optical interface between the microscope and digital camera). The images were taken at the camera’s native three TMA cores that was sufficiently stained for fractal analysis. The age range of these patients at diagnosis resolution of 1300 × 1030 pixels, and saved in tagged was 34 to 95 with a mean and median age of 65 and 66, image file format (tif). respectfully. Stage information was available for 375 of 379 patients with the following frequency distribution: Fractal Analysis to Assess Morphologic Complexity 225 (60.0%) patients were Stage I, 99 (26.4%) were Stage Unlike our intuitive notion of dimension (i.e., topologi- II, and 51 (13.6%) were Stage III. All patients selected cal dimension), fractal dimension can be a non-integer had received adjuvant tamoxifen treatment between value, and the greater the morphologic complexity of an 1988 and 2006. Cases were identified with Alberta Can- object, the higher its fractal dimension relative to its cer Board records of patients who had received tamoxi- topological dimension (Figure 1). Fractal dimension fen treatment without chemotherapy. In summary, the quantifies the level of structural complexity by assessing inclusion criterion was any patient who had adequate the variation in the level of detail in a structure as the
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 3 of 10 http://www.translational-medicine.com/content/8/1/140 Figure 1 Both the circle (left) and the Koch snowflake (right) have a topological dimension of 1; however, the fractal dimension (FD) of the Koch snowflake is greater than 1 because it has a more complex morphology than the circle. structure is examined at different scales [19]. Hence, it Figure 2 Pan-keratin stained TMA cores (left column) lends itself naturally to characterizing irregular struc- representative of A: low (< 1.56), B: intermediate (1.56-1.75), tures that maintain a constant level of complexity over a and C: high (> 1.75) fractal dimension categories, the range of scales. corresponding background corrected gray-scale images (center In this study, we applied an automated fractal analysis column), and the corresponding outline morphology images technique we developed in previous work [18] to quan- (right column) from which fractal dimensions are computed. tify the morphologic complexity of breast epithelium, a pathologically relevant histological feature. In summary, maximum corresponds to the fractal dimension of this technique involves the following steps: the pathological morphology. 1. Application of a histological stain to tissue In previous work, we showed that our method of find- specimens in order to highlight and isolate the histo- ing the fractal dimension is independent of changes in logical structures of interest. In this case, these microscope illumination setting or stain uniformity and structures include the outlines of the epithelial com- intensity [18]. Also, it should be noted that fractal ponents comprising the multi-cellular structures dimension is not affected by magnification as long as (gland formations), cellular structures (individual cell the field of view of the specimen image still contains the shapes), and sub-cellular structures (distribution of scale range of the structures of interest over which the keratin within the cells and nuclear shape). fractal dimension was found to be constant. 2. Image acquisition and background correction of Our automated fractal analysis method was applied to stained specimens. The background correction was a total of 1224 TMA cores (3 cores for each of the 408 done by acquiring a “blank” image (under the same patient samples). For each patient, the TMA core with imaging conditions used to acquire the TMA the maximum fractal dimension was used for the statis- images), and using this “blank” image to subtract the tical analysis in this study. The rationale for choosing non-uniform background luminance [18]. The the maximum fractal dimension from the sampled tissue resulting background corrected images are converted cores is to reduce the possibility that the other TMA to grey-scale (Figure 2). cores from a given patient contain only benign or more 3. Application of a series of intensity thresholds to highly differentiated tissue. That is, it is expected that convert the grey-scale version of the image specimen the TMA core with the maximum fractal dimension is into a series of binary images from which histological morphology outlines are derived (Figure 2). Figure 3 shows a sample magnified region of Figure 2A to illustrate the segmented morphology outlines in more detail. 4. Application of the box counting method [19] (with appropriate spatial scale range - 10 to 50 μm) [20] to compute the fractal dimension of each out- Figure 3 A: Original image (Figure 2C); B: Magnified portion of line image obtained from step 3. A, the dashed rectangular region; C: Segmented outline 5. Identification of the global maximum from a plot structures corresponding to the magnified image region. of fractal dimension versus intensity threshold. This
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 4 of 10 http://www.translational-medicine.com/content/8/1/140 r epresentative of the malignant neoplasm that has The prognostic accuracy of fractal dimension in pre- deviated most from normal cellular/glandular breast dicting death from breast cancer and death from any morphology, and therefore it is the most probable indi- cause was quantified by the area under the curve (AUC) cator of abnormal and/or aggressive tumor growth with from a receiver operating characteristic (ROC) analysis. metastatic potential. Values of AUC range from 0.5 (chance accuracy) to 1.0 For 379 of the 408 patients (92.9%), fractal dimension (perfect accuracy), with the following intermediate was successfully measured in at least one of the three benchmarks: 0.6 (fair), 0.7 (good), 0.8 (excellent), and TMA cores generated per patient, and it could not be 0.9 (almost perfect). For the analysis, the predicted determined for the remaining 29 patient specimens due probability of outcome from a Cox regression model to insufficient staining (i.e., less than half of the speci- was considered as a continuum. The actual occurrence men being stained) or specimen folding. Eight of the 29 of outcome was used as the comparative standard. patients could not be assessed because all 3 of their A split-sample cross-validation was performed to assess TMA cores resulted in a “blank” slide. The breakdown the generalizability of the results [21]. The process con- of the number of patients for which the TMA cores sisted of splitting the original sample of 379 patients into were sufficiently stained for fractal analysis was as fol- a training set of 190 patients and a validation set of 189 lows: 36 patients (9.5%) had one evaluable core, 105 patients using random sampling. A regression equation patients (27.7%) had two evaluable cores, and 238 was derived in the training set and the AUC between the patients (62.8%) had three evaluable cores. observed and predicted response values was calculated. The regression coefficients from the training set were then used to calculate predicted values in the validation Statistical Analyses For purposes of analyses, it is often useful to convert a set. The AUC between these predicted values and measured variable to a categorical variable so as to place observed values in the validation set was calculated, and patients into graded risk strata. As the particular fractal is called the cross-validation coefficient. The shrinkage analysis technique we developed is novel, there are no coefficient was calculated as the difference between the established cutpoints available. Although several methods AUCs of the training and validation sets. The smaller the exist to determine cutpoints, namely biological determina- shrinkage coefficient, the more confidence one can have tion, data-oriented, and outcome-oriented, there is no sin- in the generalizability of the results. Although there are gle method or criterion to specify which approach is best. no clear guidelines regarding the magnitude of shrinkage, For the present analyses, we used a data-oriented except that smaller is better, values less than 0.10 indicate approach to select two cutpoints. The first cutpoint was a generalizable model. Given a satisfactory shrinkage chosen to correspond to the upper quartile (75th percen- coefficient, the data were combined from both sets and a tile) of the fractal dimension data, and the second cutpoint final regression equation was derived based upon the was chosen as the median of the remaining lower three- entire sample. quarters of the data. Two cutpoints, rather than one, were Out of 379 evaluable patients, several had missing data: chosen to assess whether there was a graded relationship 15 (9.0%) tumor grades, 4 (1.1%) lymph node status, 15 between fractal dimension and patient prognosis. (4.0%) estrogen-receptor status, and 12 (3.2%) HER-2/ Associations between categorized fractal dimension neu status. Rather than excluding these patients from the scores and clinicopathological variables were assessed analyses and reducing the sample size, missing data were for statistical significance using a chi-square test. imputed using the predicted mean approach in SOLAS Kaplan-Meier methods were used to estimate 10-year 3.0 software (Statistical Solutions, Ltd.). Imputation bias disease-specific and overall survival rates and the log- was assessed by re-running all the analyses and excluding rank test was used to compare the curves for statistical any patient with missing data. As the estimates were significance. Disease-specific survival was measured similar, the results are reported with the imputed data. from the date of diagnosis to the date of death from Results cancer or date of last follow-up. Overall survival was measured from the date of diagnosis to the date of Fractal Analysis of the TMA Cores death from any cause or date of last follow-up. The Fractal dimension scores ranged from 1.08 to 1.97, with above analyses were repeated using Cox proportional a median of 1.62, lower quartile 1.49, and upper quartile hazards regression modeling to assess whether any of 1.75. There was moderate level of relatedness (intraclass the clinicopathological variables influenced the findings. correlation = 0.51) among the cores. Using the The proportionality assumption was assessed for all cov- data-oriented approach to select two cutpoints, fractal ariates using Log-Minus-Log Survival Plots and none dimension values < 1.56 were considered low (N = 141), violated the assumption. Statistical analyses were 1.56-1.75 as intermediate (N = 148), and > 1.75 as high performed using SAS 9.2 software (SAS Institute Inc). (N = 90). Figure 2 shows representative TMA cores
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 5 of 10 http://www.translational-medicine.com/content/8/1/140 f rom these fractal dimension categories. One can see the size of the tumour, estrogen-receptor status, or HER-2/neu status (Table 2). from this figure that the classification of TMA cores into low, intermediate, and high fractal dimension cate- gories (A-C) corresponds to the increasing complexity Tumour Grade as a Predictor of Outcome of outline morphology. Tumour grade was derived from the original pathology reports that included between 10 and 30 board-certified cancer pathologists. In contrast to the distinct separation Relationship between Fractal Dimension and Standard of the disease-specific survival curves for the different Prognosticators The baseline patient characteristics are shown in fractal dimension categories (Figure 4A), the disease-spe- Table 1. Higher fractal dimension was significantly asso- cific survival curves for grade 1 and 2 tumours virtually ciated with traditional indicators of poor prognosis, overlaped each other over the entire 10-year follow-up including older age, larger tumour sizes, higher tumour period (Figure 4C). Also, there is virtual overlap in the grade, and positive lymph node status. However, fractal overall survival curves of tumour grades 1 and 2 for the dimension was not associated with either estrogen- first 4-year period (Figure 4D). These results suggest that receptor status or HER-2/neu status. tumour grades 1 and 2 do not discriminate patients with respect to 10-year outcome. Fractal Dimension as a Predictor of Outcome The median patient follow-up was 5.2 years. The 10-year Multivariate Analysis disease-specific and overall survival rates for the entire Results from Cox proportional hazards regression group of 379 patients were 52.5% and 42.5%, respectively. showed that fractal dimension remained statistically sig- Patients with higher fractal scores had significantly worse nificant even after adjusting for all clinicopathological disease-specific survival than those with lower scores variables (Table 3). This result implies that fractal (25.0% versus 56.4% versus 69.4%, p < 0.001; Table 2 and dimension is a strong prognostic factor, even though the Figure 4A). As well, patients with higher scores had sig- multivariate hazard ratio (Table 3) is smaller than the nificantly worse overall survival (14.2% versus 39.9% ver- univariate hazard ratio (Table 2). The AUCs for the 7- sus 67.4%, p < 0.001; Table 2 and Figure 4B). The AUCs factor regression models were 0.73 and 0.75 for disease- for fractal dimension were 0.66 and 0.67 for univariate specific and overall survival, respectively. These AUCs disease-specific and overall survival, respectively, indicat- increased by only 0.07 and 0.08 when six clinical-patho- ing good levels of prognostic accuracy. As expected, logical factors were added to fractal dimension in the older age, higher grade, and positive lymph node status multivariate regression model. The small increase in were significantly predictive of worse outcome, but not AUCs incidate that the other clinical-pathological Table 1 Patient Characteristics by Fractal Dimension Category Number (%) < 1.56 (N = 141) % group 1.56 - 1.75 (N = 148) % group >1.75 (N = 90) % group P-value Age ≤ 55 years 78 (20.6) 23.4 23.7 11.1 0.039 >55 years 301 (79.4) 76.6 76.3 88.9 Size of tumour ≤ 2 cm 272 (71.8) 78.7 69.6 64.4 0.047 >2 cm 107 (28.2) 21.3 30.4 35.6 Grade of tumour 1&2 338 (89.2) 92.9 91.9 78.9 0.001 3 41 (10.8) 7.1 8.1 21.1 Lymph node status Negative 300 (79.2) 85.1 81.8 65.6 0.001 Positive 79 (20.8) 14.9 18.2 34.4 Estrogen-receptor status Positive 355 (93.7) 93.6 93.9 93.3 0.98 Negative 24( 6.3) 6.4 6.1 6.7 HER-2/neu status Negative 350 (92.4) 95.0 89.9 92.2 0.25 Positive 29 (7.6) 5.0 10.1 7.8
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 6 of 10 http://www.translational-medicine.com/content/8/1/140 Table 2 Univariate Results from Kaplan-Meier Analysis and Cox Proportional Hazards Regression Number of 10-year Disease- Univariate Hazard P-value 10-year Overall Univariate Hazard P-value Patients Specific Survival (%) Ratio (95% CI) Survival (%) Ratio (95% CI) Fractal dimension < 1.56 141 69.4 1.0 67.4 1.0 1.56 - 1.75 148 56.4 1.9 (1.1, 3.6) 0.03 39.9 2.1 (1.2, 3.6) 0.008 >1.75 90 25.0 3.5 (1.9, 6.4) < 0.001 14.2 3.6 (2.1, 6.1) < 0.001 Age ≤ 55 years 78 82.1 1.0 82.1 1.0 >55 years 301 40.8 3.3 (1.5, 7.2) 0.003 29.1 4.3 (2.0, 9.4) < 0.001 Size of tumour ≤ 2 cm 272 49.2 1.0 38.8 1.0 >2 cm 107 57.0 1.3 (0.8, 2.2) 0.21 47.9 1.3 (0.9, 2.0) 0.18 Grade of tumour 1&2 338 56.1 1.0 45.4 1.0 3 41 22.1 3.4 (2.0, 5.7) < 0.001 19.3 2.8 (1.7, 4.6) < 0.001 Lymph node status Negative 300 57.6 1.0 47.8 1.0 Positive 79 32.2 4.0 (2.5, 6.3) < 0.001 21.3 3.4 (2.3, 5.1) < 0.001 Estrogen- receptor status Positive 355 53.8 1.0 43.1 1.0 Negative 24 40.1 1.6 (0.7, 3.4) 0.26 36.0 1.6 (0.8, 3.1) 0.19 HER-2/neu status Negative 350 51.6 1.0 42.3 1.0 Positive 29 60.6 1.2 (0.5, 2.7) 0.71 38.9 1.2 (0.6, 2.5) 0.59 f actors contribute little to the prognostic accuracy generalizable and that combining data from both sets in beyond fractal dimension. It is also worth noting that the analyses was justified. even with the comparison of grades 1 and 2 as one cate- Discussion gory versus grade 3 tumours, both disease-specific and overall survival were more strongly and significantly We previously developed a fractal analysis method to associated with fractal dimension than tumour grade. quantitatively measure the morphologic complexity of epithelial architecture [18], and showed a direct associa- tion between fractal dimension and breast tumour Split-sample Cross-validation The generalizability of the aforementioned results was grade, suggesting that it may be a good surrogate mea- assessed by split-sample cross-validation as described in sure of tumour differentiation [22]. In this study we the statistical analysis section. The results, shown in examined the prognostic value of fractal dimension by Table 4 are congruent, not only with each set but also analyzing 379 specimens from patients with invasive with the results of the entire sample shown in Tables 2 breast cancer, and found that with the exception of and 3. Specifically, the frequency distribution of low, nodal status, fractal dimension showed a stronger asso- moderate, and high fractal dimension is similar, as are ciation with disease-specific survival than standard clini- the 10-year disease-specific and overall survival rates in cal prognosticators. The potential clinical implications these three categories. Even with smaller sample sizes, of these results are substantial because to our knowl- both the training and validation sets still show a pattern edge, this is the largest and only study of its kind inves- of doubling of hazards with higher levels of fractal tigating and demonstrating a positive association dimension. The shrinkage coefficients for disease-speci- between the morphologic complexity of breast epithelial fic and overall survival were -0.01 and -0.05, respec- architecture (via the fractal dimension metric) and tively, both indicating that fractal dimension is patient outcome. The potential advantages of fractal
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 7 of 10 http://www.translational-medicine.com/content/8/1/140 Figure 4 Kaplan-Meier Disease-Specific and Overall Survival Curves by Fractal Dimension Category (Panels A and B, respectively); Kaplan-Meier Disease-Specific Survival and Overall Survival Curves by Tumour Grade (Panels C and D, respectively). dimension over conventional tumour grading is that it is expectation that fractal dimension will be independent a quantitative and reproducible indicator that would be of the predictive factor related to tamoxifen therapy (i. able to provide pathologists with rapid and cost effective e., ER-positive status). Indeed, this appears to be the high volume analysis from as few as three tissue micro- case, since approximately the same percentage of ER- array (TMA) cores per patient. positive patients are in the low, intermediate, and high Ideally, a study investigating the value of a potential fractal dimension groups (Table 1), which likely indi- prognosticator should only involve patients that have cates that tamoxifen therapy has put all of these ER- not received any form of adjuvant systemic therapy. positive patients on an equal footing. However, another However, as noted by Mirza et al. [5], such studies are possibility for this result may be that ER status does becoming increasingly difficult to perform because sys- not affect the morphologic complexity of epithelial temic therapy is recommended for an ever-widening architecture. In either case, it may be argued that the range of breast cancer patients. Although none of the use of tamoxifen treated patients in a study investigat- patients in this study were treated with adjuvant che- ing the value of a possible prognosticator, although not motherapy, they were all treated with adjuvant tamoxi- ideal, does not detract from the ability to assess the prognostic factor’s potential relative to other indepen- fen therapy, including the 24 ER-negative patients (note: cases selected for this study where from as far dent prognosticators. back as 1988 when tamoxifen was occasionally admi- Previous studies have examined the application of nistered to patients with ER-negative tumours). How- fractal analysis for characterizing cancer [23,24] and ever, even though the patients received a form of have shown that fractal dimension can describe the adjuvant systemic therapy, the same form of treatment complex pathological structures seen in some cancers; was received by all of the patients leading to the [18,22] however, to our knowledge, our results represent
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 8 of 10 http://www.translational-medicine.com/content/8/1/140 Table 3 Adjusted Hazard Ratios (95% Confidence Intervals) from Cox Regression Death from Breast Cancer P-value Death from Any Cause P-value Fractal dimension < 1.56 1.0 1.0 1.56 - 1.75 1.9 (1.1, 3.5) 0.043 2.0 (1.2, 3.5) 0.011 >1.75 2.6 (1.4, 4.8) 0.002 2.7 (1.6, 4.7) < 0.001 Age ≤ 55 years 1.0 1.0 >55 years 1.8 (0.8, 4.2) 0.14 2.7 (1.2, 5.9) 0.01 Size of tumour ≤ 2 cm 1.0 1.0 >2 cm 1.0 (0.6, 1.6) 0.96 1.0 (0.7, 1.6) 0.88 Grade of tumour 1&2 1.0 1.0 3 2.1 (1.1, 3.7) 0.01 1.7 (1.1, 3.0) 0.047 Lymph node status Negative 1.0 1.0 Positive 3.1 (1.9, 5.1) < 0.001 2.6 (1.7, 4.1) < 0.001 Estrogen-receptor status Positive 1.0 1.0 Negative 1.6 (0.7, 3.7) 0.27 1.5 (0.7, 3.2) 0.26 HER-2/neu status Positive 1.0 1.0 Negative 1.1 (0.5, 2.5) 0.87 1.1 (0.5, 2.4) 0.70 the largest and sole study relating fractal dimension of demonstrates the high potential of fractal dimension as epithelial architecture to patient outcome. Although we an image-based prognostic marker, and it is congruent did not use an external patient validation set in this with the notion that malignant breast neoplasms asso- proof of principle study, we employed a data-oriented ciated with poorer outcome demonstrate partial or com- approach to minimize bias in the selection of cutpoints, plete lack of structural organization and functional as well as, conducting a split-sample cross-validation coordination with surrounding normal tissue [15]. analysis. This analysis suggests that the results are Furthermore, it implies that changes in the morphologic generalizable, whereby higher fractal dimensions are complexity of architectural components of the neoplasm associated with poorer outcome. This observation (i.e., the epithelium) that arise from changes in the Table 4 Summary of Split Sample Training Set and Validation Set Results Number of 10-year Disease-Specific Adjusted Hazard P- 10-year Overall Adjusted Hazard P- Patients Survival (%) Ratio (95% CI) value Survival (%) Ratio (95% CI) value Training Set 190 Patients Fractal dimension < 1.56 68 69.4 1.0 66.2 1.0 1.56 - 1.75 76 52.3 2.4 (0.1, 5.8) 0.064 34.2 2.2 (1.0, 4.8) 0.050 >1.75 46 17.0 2.5 (1.0, 6.3) 0.056 16.5 1.8 (0.8, 4.1) 0.17 Validation Set 189 Patients Fractal dimension < 1.56 73 71.6 1.0 70.6 1.0 1.56 - 1.75 72 60.5 1.3 (0.6, 3.3) 0.51 44.8 1.7 (0.8, 3.9) 0.18 >1.75 44 32.4 2.3 (1.0, 5.5) 0.06 11.2 3.2 (1.5, 6.9) 0.003 AUC adjusted disease-specific survival analysis, training set = 0.72, validation set = 0.73. AUC adjusted overall survival analysis, training set = 0.68, validation set = 0.73.
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 9 of 10 http://www.translational-medicine.com/content/8/1/140 functional status of cells in malignant neoplasms can be the generation of the TMA cores and database, and the interpretation of the data. All authors read and approved the final manuscript. quantified with fractal analysis. Authors’ information MT is a board certified Medical Physicist with extensive expertise in radiation Conclusions oncology physics, and medical imaging and analysis. ME is a distinguished In summary, the results of this retrospective study show Biostatistician with well over 150 publications, and expertise in the that fractal dimension is a promising image analysis mar- application of statistics to medicine. AMM is a Molecular Pathologist with ker for the prognosis of IDC of the breast. However, its’ extensive expertise in breast cancer pathology and the development and clinical implementation of prognostic and predictive molecular biomarkers prognostic value needs to be confirmed in external valida- of cancer. tion studies, and ultimately in the context of controlled Competing interests prospective clinical trials. As a step in this direction, in With the help of University Technologies International (UTI), the authors are future work, we will investigate the prognostic value of exploring the possibility of commercializing the fractal analysis software used fractal dimension for defining risk category for Stage I (i.e., to analyze the breast tissue microarray images in this study. lymph node-negative and tumour size ≤ 2 cm in maxi- Received: 20 August 2010 Accepted: 31 December 2010 mum diameter), IDC, ER-positive breast cancer patients Published: 31 December 2010 that have not received any form of adjuvant systemic ther- apy. Such a study would be especially valuable because in References 1. Lonning PE, Knappskog S, Staalesen V, Chrisanthar R, Lillehaug JR: Breast current clinical practice it is still difficult to identify this cancer prognostication and prediction in the postgenomic era. Ann subgroup of patients that would benefit most from adju- Oncol 2007, 18:1293-1306. vant chemotherapy. Also, in future work we will investi- 2. Cianfrocca M, Gradishar WJ: Controversies in the therapy of early stage breast cancer. Oncologist 2005, 10:766-779. gate the prognostic and predictive value of combining 3. Mori I, Yang Q, Kakudo K: Predictive and prognostic markers for invasive fractal dimension, a morphological index, with a quantita- breast cancer. Pathol Int 2002, 52:186-194. tive analysis of mitotic count, which is a cellular prolifera- 4. Cianfrocca M, Goldstein LJ: Prognostic and predictive factors in early- stage breast cancer. Oncologist 2004, 9:606-616. tion index that has been shown to be a significant 5. Mirza AN, Mirza NQ, Vlastos G, Singletary SE: Prognostic factors in node- prognostic indicator for node-negative breast cancer [5]. negative breast cancer: a review of studies with sample size more than These investigations would provide validation of the sig- 200 and follow-up more than 5 years. Ann Surg 2002, 235:10-26. 6. Elledge RM, McGuire WL: Prognostic factors and therapeutic decisions in nificance of morphologic complexity of epithelial architec- axillary node-negative breast cancer. Annu Rev Med 1993, 44:201-210. ture in node-negative breast cancer, and explore the 7. Polychemotherapy for early breast cancer: an overview of the randomised trials. Early Breast Cancer Trialists’ Collaborative Group. possible synergy between morphologic complexity and cel- Lancet 1998, 352:930-942. lular proliferation. Also, they will bring us closer to the 8. Tamoxifen for early breast cancer: an overview of the randomised trials. realization of an objective prognosticator that can assist Early Breast Cancer Trialists’ Collaborative Group. Lancet 1998, clinicians in making optimal treatment decisions regarding 351:1451-1467. 9. Elston CW, Ellis IO: Pathological prognostic factors in breast cancer. I. The adjuvant systemic therapy for invasive breast cancer. value of histological grade in breast cancer: experience from a large study with long-term follow-up. Histopathology 1991, 19:403-410. 10. Henson DE, Ries L, Freedman LS, Carriaga M: Relationship among Abbreviations outcome, stage of disease, and histologic grade for 22,616 cases of AUC: Area under the curve; CI: Confidence interval; ER: Estrogen receptor; breast cancer. The basis for a prognostic index. Cancer 1991, FD: Fractal dimension; H&E: Hemotoxylin and eosin; HER-2/neu: Human 68:2142-2149. epidermal growth factor receptor 2; IDC: Invasive ductal carcinoma; IRB: 11. Saez RA, McGuire WL, Clark GM: Prognostic factors in breast cancer. Semin Institutional review board; ROC: Receiver operating characteristics; tif: tagged Surg Oncol 1989, 5:102-110. image file format; TMA: Tissue microarray 12. Meyer JS, Alvarez C, Milikowski C, Olson N, Russo I, Russo J, et al: Breast carcinoma malignancy grading by Bloom-Richardson system vs. Acknowledgements proliferation index: reproducibility of grade and advantages of This work was supported by the Alberta Heritage Foundation for Medical proliferation index. Mod Pathol 2005, 18:1067-1078. Research (AHFMR) - ForeFront Block Grant. We want to thank Mie Konno 13. Robbins P, Pinder S, de Klerk N, Dawkins H, Harvey J, Sterrett G, et al: and Annie Yau for help with clinical data collection, and Chantelle Elson for Histological grading of breast carcinoma: a study of interobserver acquiring the breast specimen images. agreement. Hum Pathol 1995, 26:873-879. 14. Chowdhury N, Pai MR, Lobo FD, Kini H, Varghese R: Interobserver variation Author details in breast cancer grading: a statistical modeling approach. Anal Quant 1 Department of Oncology, University of Calgary, Calgary, Canada. 2Tom Cytol Histol 2006, 28:213-218. Baker Cancer Centre, Calgary, Canada. 3Department of Physics & Astronomy, 15. Rizki A, Bissell MJ: Homeostasis in the breast: it takes a village. Cancer Cell University of Calgary, Calgary, Canada. 4Department of Community Health 2004, 6:1-2. Science, University of Calgary, Calgary, Canada. 5Department of Pathology & 16. Coffey DS: Self-organization, complexity and chaos: the new biology for Laboratory Medicine, University of Calgary, Calgary, Canada. medicine. Nat Med 1998, 4:882-885. 17. Baish JW, Jain RK: Cancer, angiogenesis and fractals. Nat Med 1998, 4:984. Authors’ contributions 18. Tambasco M, Costello BM, Kouznetsov A, Yau A, Magliocco AM: MT performed the literature search, study design, fractal dimension analysis, Quantifying the architectural complexity of microscopic images of and drafted the manuscript and figures. ME participated in the study design, histology specimens. Micron 2009, 40:486-494. performed the statistical analysis and interpretation, and drafted the 19. Peitgen H, Jurgens H, Saupe D: Chaos and Fractals: New Frontiers of Science. statistical analysis and results sections. AM participated in the study design, 2 edition. New York: Springer-Verlag; 2004.
Tambasco et al. Journal of Translational Medicine 2010, 8:140 Page 10 of 10 http://www.translational-medicine.com/content/8/1/140 20. Dixon V, Tambasco M: Effects of image resolution and noise on estimating the fractal dimension of tissue specimens. Anal Quant Cytol Histol 2010, 32:269-279. 21. Kleinbaum DG, Kupper LL, Muller KE, Nizam A: Applied Regression Analysis and Other Multivariable Methods. 3 edition. Duxbury Press; 1998. 22. Tambasco M, Magliocco AM: Relationship between tumor grade and computed architectural complexity in breast cancer specimens. Hum Pathol 2008, 39:740-746. 23. Baish JW, Jain KJ: Fractals and cancer. Cancer Res 2000, 60:3683-3688. 24. Cross SS: Fractals in pathology. Journal of Pathology 1997, 182:1-8. doi:10.1186/1479-5876-8-140 Cite this article as: Tambasco et al.: Morphologic complexity of epithelial architecture for predicting invasive breast cancer survival. Journal of Translational Medicine 2010 8:140. Submit your next manuscript to BioMed Central and take full advantage of: • Convenient online submission • Thorough peer review • No space constraints or color ﬁgure charges • Immediate publication on acceptance • Inclusion in PubMed, CAS, Scopus and Google Scholar • Research which is freely available for redistribution Submit your manuscript at www.biomedcentral.com/submit